Prefect ops

Skill bgevorkian/agent-skills/skills/prefect-ops

Reusable open-source Agent Skills for Pi and compatible agent systems

Install

npx -y skills add bgevorkian/agent-skills --skill prefect-ops

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

13 days oldThe repository was created 13 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Inspect and manage a Prefect 3 API from a safe JSON CLI. Use when the user explicitly asks to read Prefect runs, deployments, logs, variables, automations, work pools, or to perform a confirmed operational change such as running a deployment or cancelling a flow run.

The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.

SKILL.md

3.9 KB, as published. Nobody here has run it

Prefect Ops

Generic Prefect 3 REST skill. The helper uses only the Python standard library, prints UTF-8 JSON to stdout, avoids echoing secrets, and keeps reads and writes clearly separated.

Configuration

Provide the API endpoint through an environment variable or flag:

Variable	Flag	Required	Notes
`PREFECT_API_URL`	`--api-url`	yes	Prefect API base URL, for example `https://prefect.example/api`
`PREFECT_API_KEY`	none	no	Optional bearer token; environment-only to avoid shell history/process-list exposure
`PREFECT_ALLOW_WRITE`	none	for writes only	Must be exactly `true` to enable mutations

Writes require both:

PREFECT_ALLOW_WRITE=true
--confirm-write

Reads never require either gate.

Secret setup

Before configuring credentials, ask which secret manager and local profile the user wants. Follow Secure secret profiles. Do not invent or publish profile names, hosts, templates, or secret references. If the user asks for the author's method, use a per-profile Proton Pass pointer file with process-scoped pass-cli run. Never request or display resolved values.

Run

From this skill directory:

uv run --python 3.13 python scripts/pf.py --api-url https://prefect.example/api server-version
uv run --python 3.13 python scripts/pf.py flow-runs --limit 20
uv run --python 3.13 python scripts/pf.py deployments --limit 50
uv run --python 3.13 python scripts/pf.py deployment --deployment my-flow/my-deployment
uv run --python 3.13 python scripts/pf.py logs --flow-run 00000000-0000-0000-0000-000000000000 --limit 100
uv run --python 3.13 python scripts/pf.py scheduled-runs --deployment my-deployment --limit 10
uv run --python 3.13 python scripts/pf.py variables --limit 100

Write examples:

PREFECT_ALLOW_WRITE=true uv run --python 3.13 python scripts/pf.py run --deployment my-flow/my-deployment --param retries=2 --confirm-write
PREFECT_ALLOW_WRITE=true uv run --python 3.13 python scripts/pf.py cancel --id 00000000-0000-0000-0000-000000000000 --confirm-write
PREFECT_ALLOW_WRITE=true uv run --python 3.13 python scripts/pf.py pause --deployment my-flow/my-deployment --confirm-write
PREFECT_ALLOW_WRITE=true uv run --python 3.13 python scripts/pf.py variable-set --name feature_flag --value true --confirm-write

--deployment accepts:

a deployment UUID
flow_name/deployment_name
a bare deployment name, if it resolves to exactly one deployment

Read commands

flow-runs
flow-run
task-runs
logs
deployments
deployment
schedules
scheduled-runs
variables
variable
automations
automation
work-pools
server-version

Write commands

run
cancel
retry
delete
pause
resume
set-state
variable-set
variable-delete

Safety contract

API keys are never printed.
Every write is double-gated by environment and CLI confirmation.
Response sizes, timeouts, and list limits are bounded.
HTTP failures return concise JSON errors on stderr.
Deployment lookup is generic and contains no bundled names or infrastructure assumptions.
Prefer a server-side read-only or least-privilege token when possible.

Tests

From the repository root:

uv run --python 3.13 python skills/prefect-ops/tests/test_pf.py
uv run --python 3.13 scripts/validate_skills.py